Cross-Robot-Knowledge-Transfer-Lite Kinova Gen3 Lite Dataset experiments
Python 3.10 and MATLAB R2022b are used for development.
git clone https://github.com/gtatiya/paper5.git
cd paper5
pip install -e .
Install MATLAB Engine API for Python
MATLAB Dependencies: Statistics and Machine Learning Toolbox
- Download the dataset and create a symbolic link to it as
data:
Windows: mklink "data" "<path to dataset>"
Linux: ln -s "<path to dataset>" "data"
- Create dataset:
python data_processing/create_dataset.py - Discretize data:
python data_processing/discretize_data.py - Autoencode data:
python data_processing/autoencode_data.py - Plot data:
python data_processing/plot_data.py - Plot features:
python data_processing/plot_features.py
- Learn object recognition:
python learn/classify_objects.py - Learn object recognition:
python learn/classify_objects_v2.py - Learn tool recognition:
python learn/classify_tools.py
- Transfer robot knowledge:
python transfer/robot_knowledge.py -increment-train-objects -num-folds 10 -augment-trials 10
python transfer/robot_knowledge.py -increment-train-objects -augment-trials 10 -across tools -feature autoencoder-linear-tl
- Analyze results and plots:
python analyze/transfer_results.py - Plot KEMA features:
python transfer/plot_kema_features.py